Tackling Systems and Security Challenges of Edge Computing Environments

Modern applications like augmented reality (AR), virtual reality (VR), autonomous vehicles, and smart city systems require low-latency data processing. However, the devices on which these applications run often have limited computational resources. To address this issue, edge computing has been introduced, where resource-constrained devices offload computation tasks to external computing resources located physically close to the end-user, rather than to a remote cloud. On the other hand, an edge computing network acts like a small data center with constrained computational capability and hardware resources, and how to make full use of these resources is crucial to support a large number of users without deploying more infrastructure. One common critical observation in many edge computing scenarios is that a large number of user devices are often in close proximity and frequently offload tasks with similar input data, resulting in identical final outputs. As a result, the output of a task can be stored and reused later for tasks with similar input (i.e., computation reuse), instead of executing them from scratch. In this dissertation, we propose a framework called Reservoir for distributed edge computing environments, where tasks with similar input data are forwarded to the same computing node (server) to maximize the opportunity for computation reuse. We also introduce Deduplicator, a middleware system that achieves load balancing among edge servers while facilitating computation reuse. Besides resource efficiency, security, and privacy are major concerns in edge computing. Offloading computations to untrusted environments over the network can expose sensitive information, and malicious actors can potentially infer the nature of computations and orchestrate targeted attacks. To address these challenges, we propose a framework, Camouflage, that allows users to anonymize offloaded computations so that it is hard (if not impossible) for attackers to distinguish the tasks running on edge servers. This dissertation provides a comprehensive investigation of improving edge computing efficiency, scalability, and security by leveraging computation reuse and anonymizing offloaded tasks to enable next-generation latency-sensitive applications.